Selected Works
A collection of experiments, tools, and research.
July - Ongoing, 2024
lostdb - A fast and persistent key value store
PythonDatabase Systems
- Designed a key value database from scratch in Python with a disk-based persistence layer, enabling 100% data recovery and consistency between restarts.
- Implemented a B-Tree indexing structure, reducing disk reads by 50% and improving query performance.
- Developed serialization and deserialization methods for database nodes, converting data to and from the slotted page format for minimal overhead and efficient access.
September, 2022
Building and Road Segmentation using UNet and Transfer Learning Approach
Deep LearningPyTorchNumpyMatplotlib
- Developed a novel EffUNet architecture combining EfficientNetV2 as an encoder with a UNet decoder for segmenting buildings and roads from aerial images, achieving benchmark mIOU scores of 0.8365 and 0.9153.
- Utilized PyTorch, Numpy, and Matplotlib to implement and evaluate the model, demonstrating superior performance with V2L+UNet architecture, achieving high precision and recall rates.
- Conducted comprehensive model evaluations, optimizing for metrics such as Dice Loss, F1 Score, and accuracy, contributing valuable data for urban planning and policy-making.
May 2021Research Paper
LSB Steganography Using Pixel Locator Sequence with AES
PythonPillowOpenCVCryptography
- Developed and published an advanced LSB steganography technique using a Pixel Locator Sequence (PLS) and AES encryption, enhancing data security by randomly distributing pixel modifications.
- Implemented and evaluated the technique with Python, Pillow, OpenCV, and Cryptography libraries, achieving robust steganalysis results, such as 0.0365 (Primary Sets) and 0.0203 (Chi-Square).
- Published findings in IEEE ICSCCC'21, showcasing significant improvements in steganography security and practical applications in confidential data embedding and transmission.
Read Paper
Presented at ICSCCC'21, NITJ
January, 2022
Saathi | AI based agriculture utility website
Web DevelopmentMLTensorflowKerasFlaskBootstrapSQLDockerNginx
- Developed and deployed an AI-powered agriculture platform using TensorFlow, Keras, and Scikit-learn, achieving up to 99.31% accuracy in crop recommendation and over 95% accuracy in plant disease detection.
- Built a user-friendly web interface with Flask and Bootstrap, providing comprehensive crop information and tailored recommendations based on soil quality and environmental factors.
- Containerized the application environment with Docker and implemented Nginx for load balancing, enhancing deployment consistency, scalability, and website performance under high traffic conditions.
December, 2021
Image Quilting - Texture generation
Digital Image ProcessingPython
- Implemented the paper of Alexei A. Efros & William T. Freeman on image quilting texture generation and transfer from SIGGRAPH'01.
- Replicated the results by implementing it in Python using Numpy, OpenCV and Matplotlib libraries.
October, 2021
bWall | Anonymous Image-based bulletin board
Web DevelopmentFlaskBootstrapSQLDocker
- Developed an anonymous image-based bulletin board platform where users can upload images, comment and share content.
- Implemented backend with Flask and performed comprehensive backend unit tests.
Summer'21
Postgres - Auto Indexing
Database SystemsCPython
- Automated the indexing for secondary columns with large scan read requests by Postgres to increase efficiency.
- Studied workings of Postgres, used HypoPG for virtual indexing to avoid overhead by parsing the query information to a log file.
May 2021 - June 2021
Game of Life
Theory of AutomationC++
- Created a 'Game of Life' simulator in C++ which is a cellular automaton devised by John Conway.
- Implemented this simulation system in C++ using functional programming paradigm.
December, 2020
CovidRecognizer
Image processingDeep LearningOpenCVNumPyPillow
- Created an image processing pipeline for biomedical image classification and segmentation.
- Used image processing techniques such as CLAHE, binarization, dilation, etc.
- Implemented UNet for segmentation and achieved an accuracy of 82%.
May, 2020
Covid19 Tracker
Web DevelopmentHTMLCSSBootstrapJavaScript
- Displayed the covid infection, recoveries and deaths of every state in India.
- Fetched data through API and displayed the results in a bar format using Chart.Js.